No, Instant Runoff wouldn’t solve spoiled elections.

Here’s what would. And why you haven’t heard of it.

A bad voting system, a bad leader.

In 2012, Egypt was gearing up for a historic moment: freely electing a president for the first time ever. There were several candidates, but three basic paths forward: the old guard, the Muslim Brotherhood, or non-Islamist reformers. They chose what they thought would be a spoiler-proof election system: two-round runoff.

And they got it wrong. The winner of the election, Mohamed Morsi of the Muslim Brotherhood, overreached his mandate, became widely unpopular, and was deposed by the army the next year. The 2012 election was not only the first free presidential election in Egypt, it may also be the last, at least for some time.

This was not a failure of democracy itself, but a failure of a specific voting system. Even at the moment he was elected, Morsi was not favored by a majority of Egyptians, and polls suggested that he would have lost in a one-on-one election against a reformer such as Amr Moussa or Abdel Aboul-Fotouh.

To understand what went wrong, we have to look at the concepts of spoiler candidates and spoiled elections. A spoiler candidate is someone who does not win an election, but who changes the result by participating. In plurality voting — the system used in the US—it’s easy to see how that could happen. The classic example is Ralph Nader in Florida 2000; if Nader had not run in Florida, and if his voters had gone to Gore by even a tiny fraction more than they went to Bush, then Gore would have won the state.

(Of course, given that the margin was so close, there were plenty of other people and things that can also be blamed for Bush’s win. But that’s irrelevant here. A car needs gas AND wheels AND a distributor cap to run; and it turned out that Bush needed Nader AND Katherine Harris AND the Supreme Court to win Florida. Nader does not have to be a sufficient cause to be a spoiler, just a necessary one.)

In plurality voting, spoilers are nearly inevitable as soon as there are more than 2 significant candidates in a race. Even if the third candidate gets less than 3%, as Nader did, they can be a spoiler if the other two are close to evenly-balanced. Luckily, Egypt realized that, and didn’t use plurality. Unfortunately, the system they used, a two-round runoff, is also subject to spoilers.

How can that be possible? Only the two strongest candidates from the first round, Morsi (Muslim Brotherhood) and Ahmed Shafik (old guard), were even competing in the second round, so that round cannot be spoiled. But the problem is that runoffs tend to prematurely eliminate centrists after the first round. Extremists tend to be either first choice or last choice; but for centrists, a significant part of their true support is in the form of second-choice preferences; either side would support them against the other. Since runoff systems ignore that second-choice support, they eliminate the centrist, even if that person could beat either of the extremes in a one-on-one race. So in the Egyptian election, Shafik was likely a spoiler; unable to beat candidates like Moussa or Sabahi, he nonetheless knocked them out in the first round.

Egypt was using a two-round runoff. But this same thing would have happened in an Instant Runoff (IRV) election. In this kind of election, voters rank the candidates in order of preference, then candidates are eliminated one-by-one starting from whichever is most preferred (among those who remain) by the fewest voters. This means that every time a candidate is eliminated, their votes are redistributed to the next preference of their voters, or discarded if the voter did not specify a next preference. Thus, when only two candidates remain, one of them will necessarily have a majority of the remaining un-discarded ballots (though this may not be a majority of the original ballots.) It’s entirely possible that IRV would have gotten the same result as the two-round system did in Egypt.

That’s not to say that IRV wouldn’t help avoid spoilers sometimes; just that there are still situations where they’re inevitable without the same lesser-evil voting and two-party domination that people hate about plurality. Imagine Florida 2000 had been run with IRV. In this world, Nader would have gotten more support from people who, in the real world, cast strategic votes for Gore. If Nader’s resulting total were 20%, he’d be eliminated, his votes would be redistributed, and the true majority winner between Bush and Gore would win; IRV would have avoided a spoiled election. But if Nader’s total were 26%, that would leave Gore with a lower total, so Gore would be eliminated. His votes would be redistributed, some would go to Bush, and Bush would likely win, even if Gore could have beaten him one-on-one. In that case, IRV would have allowed Nader to be a spoiler; and yet Nader would surely argue that, as second-place finisher, he couldn’t possibly be a spoiler. We’d be back to the same circular high-stakes arguments between purists and pragmatists that voting reform should be able to avoid.

I’ve only mentioned two elections so far — Egypt 2012 and USA 2000—but it would be hard to calculate the enormous damage done by these two spoiled results. Elections matter, and second- or third-rate election systems like IRV or Plurality can be immensely harmful.

There are systems which solve this problem. In Approval Voting, you can vote for as many candidates as you want, and the one with the most votes win. That means there’s always a way to vote for your favorite candidate and still avoid spoiling the election; just make sure to approve at least one viable candidate. All candidates would show their true level of support, and centrists could still win. (Approval voting is the system supported by electology.org, a nonprofit on whose board I serve.)

If that’s too much compromising for you, then there’s Majority Judgment, where you grade each candidate on a scale such as A-F, tally their votes from the top grade down, and elect the first one to reach 50% (highest median grade). Though that’s a bit more complicated than Approval, it is even easier for voters to avoid spoiled elections. In fact, you could be almost certain to vote optimally without worrying about which candidates are viable, only grading them honestly across the available grades.

So if there’s good systems, why am I even talking about Instant Runoff? Because if you’ve heard of just one voting system reform proposal, it’s probably IRV. The reason for that gets into a bit of inside baseball, and in the US, it centers on one organization: FairVote. (See the letters “IRV” in the middle of their name there?)

FairVote was born in the 90s, and it grew out of an organization called Citizens for Proportional Representation (CPR). Proportional representation (PR) systems can help fix the problem of gerrymandering by proportionally electing multiple winners at once to a legislature. PR is great and I’ll discuss it a bit more below, but the important thing is that the most well-known form of PR, especially back in the 90s, was called Single Transferrable Vote (STV), and IRV is what you get if you use STV to elect just a single winner.

So FairVote decided to push for IRV, partly as a stepping stone towards STV. But in doing so, they ignored a number of defects that arise when you successively eliminate candidates all the way down until just one remains:

  • As discussed above, IRV can prematurely eliminate centrists, similar to the Egyptian result. This actually happened with IRV in Burlington, VT, 2009, in their first election using IRV; since this kind of result is by definition opposed by a majority of voters, it’s no surprise that they ended up repealing IRV, even though it’s tragic that they went back to using plurality which is even worse.
  • IRV requires centralized counting, making it completely unworkable for US presidential elections. (Approval, on the other hand, could be used on a state-by-state level, or even nationally by means of an interstate compact involving a majority of electoral college votes.)
  • IRV is “nonmonotonic” and statistically unstable. This sounds technical but you can see the results in these visualizations developed by Ka-Ping Yee:
IRV results in two-dimensional “issue space”. Each dot is a candidate, each pixel a different election where the voters are centered around that pixel. The crazy shatter of colors around the center shows nonmonotonicity. Note that even when Yellow is at the center of all voters, green wins. See the link above for more explanation and more diagrams like this.
As above for plurality voting. No crazy monotonicity, but now Blue and Green have a huge unfair advantage.
As above for approval voting. Whichever candidate is closest to the center, wins. Period.
  • For the reasons above, IRV is hard to audit or recount.
  • It’s hard for some voters to rank all the candidates, which may depress and bias turnout and which leads to additional invalid ballots.
  • It’s hard to explain. Even in places that use it, misconceptions abound.

In 2000, I was a supporter of FairVote; I even made a web page with an FAQ in support of IRV. But as time went on, I and others began to understand the above problems better. And we realized that approval voting had none of those problems. Approval voting won in a 2011 poll of voting experts at the London School of Economics; and a group of amateur activists came together to sign a declaration supporting several possible systems including approval and majority judgment but not IRV.

But as the hits kept coming, FairVote just kept doubling down on IRV. They campaigned for IRV in places that already used runoffs, on the argument that it would save money by avoiding extra elections; however, given the extra costs of IRV-compatible machines and voter education, this savings is very hard to see, and on a theoretical basis, IRV is arguably inferior to two-round runoffs. Thus, cities like San Francisco, Oakland, and Minneapolis have been left worse off than they were before (though still better than cities using plurality without runoffs).

Activists like me began to question the FairVote leaders. They responded with one true argument and several false ones.

The true argument is the one legitimate advantage of IRV: it obeys the “later no-harm” criterion, meaning it never hurts to add additional preferences to your ballot. This, they claimed, would mean voters would rank all relevant candidates, leading to good results. IRV is just about the only system which meets this criterion, though Majority Judgment meets a slightly weakened version of it (as long as you grade later preferences below the winning median).

But along with this truth, they persisted in misrepresenting the system in several ways, even after it had been pointed out. They suggested that IRV guarantees a majority winner; as Oakland results show, because some voters don’t fully rank candidates, it doesn’t. They suggested that approval voting would strategically devolve to plurality because voters would “strategically” just vote for one candidate (“bullet vote”); this is untrue from both a theoretical and empirical perspective. In blatant contradiction with the previous point, they also argued that approval could fail to elect a candidate preferred by an overwhelming majority; this is a theoretical possibility, but again both game theory and experience argue that it will never happen, and it actually couldn’t happen if voters bullet voted.

Recently, FairVote even began a push to rebrand IRV as “RCV”, for Ranked Choice Voting. This is infuriating to voting theorists, because that is an existing umbrella term for systems including IRV, Condorcet, and Borda, but excluding systems like Approval, Majority Judgment, and Score. Thus the term is simultaneously too broad to refer to just IRV, and too narrow to refer to voting reform in general.

After these debates had been going on for years, a group of activists decided to form a new organization, electology.org, of which I’m now a board member. We’re open to any voting system that can show it will work well at actually satisfying voters. To see which systems will do that, we’ve helped refine a clear metric known as Voter Satisfaction Efficiency (formerly, Bayesian Regret), based on monte-carlo simulations of virtual voters. This metric suggests Approval is far superior to IRV.

We’re still a lot smaller than FairVote, but we’ve been growing quickly. We’ve helped prize-giving organizations like the Webbies and the Hugo Awards develop and implement better voting systems to recognize prizeworthy candidates; we’ve helped craft approval voting bills for several states such as NH, AZ, and CO (though none have passed yet); we’ve talked to political parties and convinced several state-level bodies to use approval voting to choose candidates; and we’ve been working to educate people about voting systems.

We still agree with FairVote on several issues. Most of all, we share a recognition that plurality voting and single-member districts is just about the worst voting system possible, and that reform is an important, win-win option. We support their push for proportional representation, and think STV is one good PR system (though we also like other systems, such as at-large open-list PR with district-specific ballots, or a German-style mixed-member open-list system).

If this area is interesting to you, you should check out our website, subscribe to our newsletter, and consider donating. And if you’ve learned something from this article, please respond and/or click the heart. Thanks for reading!